Are ETFs Replacing Index Mutual Funds? · mutual funds and propose a “fair pricing” mechanism...
Transcript of Are ETFs Replacing Index Mutual Funds? · mutual funds and propose a “fair pricing” mechanism...
Are ETFs Replacing Index Mutual Funds?
Ilan Guedj and Jennifer Huang∗
March 2010
Abstract
We develop an equilibrium model to investigate whether an Exchange-Traded Fund
(ETF) is a more efficient indexing vehicle than an Open-Ended Mutual Fund (OEF).
We find that while flow-induced trading is costly to OEF investors, it is also beneficial
to those investors who cause the flow – it is simply a zero-sum game. Indeed, the OEF
structure can be viewed as providing insurance for investors with liquidity shocks, and
hence is beneficial for risk averse investors. However, this liquidity insurance is not
without cost – it can cause moral hazard and reduce the OEF performance. Moreover,
we find that investors with higher individual liquidity needs prefer to invest via the OEF
since they benefit more from the liquidity insurance. Surprisingly, the OEF structure
is still viable despite the concentration of higher-liquidity-need investors in the OEF.
The reason is that flow-induced trading costs depend only on the aggregate liquidity
need, not on individual liquidity needs, which cancel out at the fund level. As a result,
OEFs and ETFs coexist in equilibrium with different liquidity clienteles. Finally, we
derive empirical predictions that ETFs are better suited for narrower and less liquid
underlying indexes, and for investors with longer investment horizons.
∗McCombs School of Business, University of Texas at Austin. Guedj: [email protected] and (512)471-5781. Huang: [email protected] and (512) 232-9375. The authors thank FranklinAllen, Daniel Bergstresser (discussant), Marshall Blume, Tarun Chordia, Douglas Diamond, Darrell Duffie,William Goetzmann, Wei Jiang (discussant), David Musto (discussant), Francisco Perez-Gonzalez, OlegRytchkov, Jacob Sagi, Gus Sauter (discussant), Clemens Sialm, Paolo Sodini, Johan Sulaeman (discussant),Sheridan Titman, Charles Trzcinka (discussant), Luis Viceira, participants at the Institutional InvestorsConference at UT-Austin, the Lone Star Conference, 19th Annual Conference on Financial Economics andAccounting, 2009 American Finance Association annual meeting, and 38th Wharton Conference on House-hold Portfolio-Choice and Financial Decision-Making, and seminars at BI Norwegian School of Managementand Stockholm School of Economics for comments and suggestions. Support from Q-group is gratefullyacknowledged.
1 Introduction
Since the introduction of the first Exchange-Traded Fund (ETF) in the U.S. in 1993, ETFs
have captured most of the growth in index mutual funds and now constitute about 40% of the
index fund market share. Despite their growth, not all practitioners are convinced of their
value as indexing tools. John Bogle, founder of Vanguard, has been the most vocal critic of
ETFs: “if long-term investing was the paradigm for the classic index fund, trading ETFs can
only be described as short-term speculation.”1 On the other hand, Lee Kranefuss, CEO of
iShares at Barclays Global Investors, argues that ETFs are to mutual funds what compact
discs were to records: “a better product with more features for less money,” especially since
“long-term ETF investors don’t subsidize the costs of active traders in ETFs.”2
The view that ETFs are a more efficient indexing vehicle is rooted in the fact that fund
flows to an Open-Ended Index Mutual Fund (OEF) can be costly. In particular, whenever
OEF investors purchase or redeem shares, their demand is pooled at the fund level and
transacted at the daily closing price of the fund, which does not fully account for the future
price impact that the fund may experience in implementing the pooled demand. Thus, there
is cross-subsidization among investors – either existing investors subsidize new investors in
the case of net fund inflows or the remaining investors subsidize the departing investors in
the case of net fund outflows. This is the cost of the OEF structure that Kranefuss and
other ETF advocates have emphasized. This cost is also well recognized in the academic
literature. For example, Edelen (1999) documents that the extra trading (and the resulting
price impact) induced by fund flows can reduce the average performance of a fund by as
much as 1.4% per year.3 Greene and Hodges (2002) find that even daily fund flows, which
are more transient and less likely to cause mutual fund managers to incur transaction costs,
can impose significant costs on the fund, especially for funds with large daily flows. Dickson,
Shoven, and Sialm (2000) demonstrate that fund flows negatively affect mutual funds’ after-
tax returns. Johnson (2004) shows that different types of investors impose different flow
costs on the fund, and Christoffersen, Keim, and Musto (2007) find that index funds have
even higher trading costs than active funds – despite the lack of information content of
1“‘Value’ Strategies”, WSJ, John Bogle, February 9, 2007.
2See Viceira and Wagonfeld (2007) for a detailed description of BGI’s success in the ETF industry.
3Gastineau (2004) applies Edelen (1999)’s cost estimation and concludes that the cost of providing liq-uidity for open-ended mutual funds can be as high as $40 billion per year.
1
their trades – due to the inflexibility of index funds in meeting liquidity needs.4 Overall,
this literature points to significant costs of the OEF structure associated with flow-induced
trading.5
ETFs, in contrast, are designed not to have flow-induced trading costs. Like closed-ended
index mutual funds (CEFs), ETFs are traded on the stock exchange with individual traders
bearing their own transaction costs. The main difference from the CEFs is that ETFs allow
some creation and redemption of shares via the in-kind redemption feature. That is, some
authorized investors can exchange the underlying index assets for shares of the ETF or vice
versa.6 This feature allows these investors to arbitrage away any price deviation between
the ETF and the underlying index assets. While the in-kind redemption is similar to the
creation or redemption of shares via fund flows in OEFs, it does not lead to flow-induced
trading costs, since ETFs receive (or pay out) the underlying index assets directly. There is
no need to purchase (or liquidate) the underlying assets or to incur any related costs.
While ETF investors can avoid the flow-induced trading costs at the fund level, they need
to pay their own transaction costs when purchasing or liquidating their shares. It is therefore
not clear whether on average investors are better off investing in ETFs and whether ETFs
are a more efficient indexing vehicle than OEFs. In this paper, we develop an equilibrium
model to compare the efficiency of ETFs and OEFs for small investors with various individual
liquidity needs. We ask whether the rapid growth of the ETF industry implies a permanent
structural shift in the mutual fund industry.
Our first result is that ETFs are not more efficient than OEFs – it is a zero-sum game
between those who cause the fund flow and those who bear the flow-induced trading costs.
4There is a large literature documenting the large transaction costs incurred by actively managed funds.Grinblatt and Titman (1989) find the total costs to be as high as 2.5% per year. More recently, Edelen,Evans, and Kadlec (2007) find that the annual trading costs for a large sample of equity funds are comparablein magnitude to their expense ratio. These costs are higher for larger funds, especially if the fund has largerrelative trade sizes, suggesting that trading costs contribute to the diseconomies of scale for mutual fundsidentified in Chen, Hong, Huang, and Kubik (2004).
5Sophisticated investors may even exploit the pricing scheme of mutual funds to design profitable tradingstrategies. For example, Chalmers, Edelen, and Kadlec (2001) find that on extreme days the failure toaccount for nonsynchronous trading in determining funds’ net asset value can result in an average one-day excess return of 0.84 percent at high beta small-cap domestic equity funds. Goetzmann, Ivkovic, andRouwenhorst (2001) show that the issue of nonsynchronous trading is most pronounced in internationalmutual funds and propose a “fair pricing” mechanism that partially corrects net asset values for stale prices.Zitzewitz (2006) provides evidence that some investors may even abuse this pricing feature by trading after4pm, and that such late trading can lead to significant shareholder welfare loss.
6Only authorized participants can perform in-kind redemption. See Section 2 for detailed description.
2
Indeed, the OEF structure can be viewed as providing partial insurance against future liquid-
ity needs: Investors pay the cost of lower average returns in exchange for better prices when
they experience liquidity shocks.7 As long as investors are risk averse, the OEF structure is
actually beneficial rather than costly to investors. However, this benefit may not be obvious
to the investors, since the reported performance of OEFs can be lower than that of ETFs
due to the difference in the way returns are accounted for. In particular, the flow-induced
transaction costs are equally shared by all investors in the OEF and reduce reported OEF
performance, whereas they are incurred only by those ETF investors with liquidity needs
and do not affect reported ETF performance.
Second, we find that the liquidity insurance embedded in the OEF structure is not without
cost – it can cause moral hazard issues that induce excessive trading and reduce the OEF
performance. In particular, since OEF investors transact at a price that does not fully
incorporate the impact of their own trading decisions, they do not internalize the price
impact of their orders. As a result, they trade too aggressively given their trading needs,
leading to excess trading costs for the OEF. This extra trading cost makes the liquidity
insurance an ex-ante negative-sum game and leads to a lower average return for the OEF.
Third, the tradeoff between the liquidity-insurance benefit and the moral hazard cost
differs for investors with different individual liquidity needs. Under the OEF structure, the
moral hazard cost of excessive trading is shared evenly among all investors. Higher-liquidity-
need investors benefit more from the liquidity insurance and prefer to invest via the OEFs,
whereas lower-liquidity-need investors benefit less from the liquidity insurance and prefer to
invest via the ETFs. As intuitive as it may seem, this result runs directly counter to John
Bogle’s critique that “ETFs can only be described as short-term speculation.” Instead, we
find that long-term investors may optimally choose to invest in the ETFs and that there is
a viable role for ETFs in the mutual fund industry.
One might be concerned about the viability of the OEF structure given the concentration
of higher-liquidity-need investors in it. Our fourth result states that the OEF structure is
still viable and that OEFs and ETFs coexist in equilibrium with different liquidity clienteles.
The reason behind this rather surprising result is that flow-induced trading costs depend only
on the aggregate liquidity need at the fund level. Since individual liquidity needs cancel out
across investors, the concentration of investors with high individual liquidity needs does not
7This is similar to the insurance feature of bank deposit contracts in Diamond and Dybvig (1983).
3
lead to higher aggregate liquidity need or higher flow-induced trading costs for the OEF.
This result highlights a potential pitfall of the recent trend in the mutual fund industry
to impose frequent-trading restrictions. Since higher frequency traders do not necessarily
impose higher costs to the OEF yet they benefit much more from the liquidity insurance
provided by the OEF structure, they are the best clientele for OEFs. If they are deterred
from investing in the OEFs by frequent trading restrictions, OEFs may lose a significant
client base.8
Finally, our model provides a framework to assess the effectiveness of the OEF and the
ETF structures under different circumstances and to make predictions regarding the long-
term trend in the growth of the mutual fund industry. One prediction of the model is
that if investors have more correlated liquidity needs, the OEF is expected to have larger
unbalanced demand from its investors. The price impact is higher, making the OEF less
attractive relative to the ETF as an indexing vehicle. As a result, we expect to see a smaller
size of the OEF industry in equilibrium. This prediction is actually verifiable. For example,
it is reasonable to assume that investors in a narrower index (such as a bio-tech index) are
more likely to have correlated liquidity needs compared to investors in a broader market
index (such as the S&P 500 index). Hence, we expect to have more ETFs in narrower
indexes than in broader market indexes. Similarly, less liquid underlying indexes are likely
to generate a higher price impact and larger tracking error for OEFs, making them less ideal
candidates for OEF investing as well. These patterns are largely consistent with the growth
pattern of ETFs in the market: Many new ETFs track indexes that have less number of
stocks, high industry concentration, high volatility, and low liquidity.
Despite the phenomenal growth of the ETF industry, there is limited academic research on
ETFs, mostly due to this sector’s short history. Elton, Gruber, Comer, and Li (2002) study
SPDR, the first ETF that tracks the S&P 500 index, and document that it underperforms
relative to both the underlying index and to other OEFs tracking the same index. Poterba
and Shoven (2002) look at the tax implications of ETFs in general and the performance of
SPDR in particular. They find that although in theory ETFs can be more tax efficient, in
reality SPDR ETF performs slightly worse than the Vanguard S&P 500 both in before-tax
8This result is based on the assumption that higher-liquidity-need investors do not have higher exposureto the systematic liquidity risk. If there is reason to believe that frequent traders are more likely to bereacting to market conditions or are more correlated in their trading decisions, then they may impose highercosts to OEFs and some forms of frequent trading restrictions may be optimal.
4
and after-tax returns. We add to this literature by placing ETFs in a broader context,
to understand their impact on the structure of the mutual fund industry. Agapova (2006)
compares fund flows into index mutual funds and ETFs and finds that their coexistence can
be partially explained by a clientele effect.
Our paper is most related to Chordia (1996) and Chen, Goldstein, and Jiang (2007),
who consider the cross-subsidization induced by the pricing scheme of the OEF structure.
Chordia (1996) is the first to point out that this cross-subsidization can be viewed as a form
of insurance against liquidity shocks along the line of Diamond and Dybvig (1983). His
focus, however, is quite different. He shows that funds optimally hold more cash in order
to meet redemption demands and that load and redemption fees can help alleviate liquidity
impacts. Chen, Goldstein, and Jiang (2007) show that the cost of this cross-subsidization can
manifest into a “bank run” in which investors flee the fund for fear of bearing the liquidity
cost imposed by others’ withdrawing from the fund. We show that both features are present:
While the liquidity insurance feature embedded in OEFs is beneficial, especially for higher-
liquidity-need investors, it can induce moral hazard issues and reduce the performance of
ETFs. As a result, lower-liquidity-need investors may prefer the ETF structure.
Another related literature investigates the interaction between investor behavior and the
organizational structure of mutual funds. Nanda, Narayanan, and Warther (2000) show
that the existence of investor clienteles with differing liquidity and marketing needs gives
rise to a variety of open-ended fund structures that differ in the average return delivered to
investors. Massa (1997) suggests that the vast number of funds offered to investors can be
seen as marketing strategies used to exploit investor heterogeneity and that market forces
may induce a sub-optimal number of mutual funds and categories. Christoffersen and Musto
(2002) show that the price sensitivity of individual investors affects the pricing scheme of
mutual funds. Our paper complements this literature by introducing a new investment
vehicle (an ETF) with a different trading mechanism and studying investor choice between
the new and the existing vehicles in equilibrium.
This paper also relates to the literature on the structure of the mutual fund industry. The
efficiency of the open-ended structure has been much debated. As discussed earlier, a large
literature documents the cost of the OEF structure associated with flow-induced trading.
Separately, a smaller, mainly theoretical literature, suggests a potential benefit of the OEF
structure when managers have ability. For example, Berk and Green (2004) show that when
5
investors vote with their feet in open-end mutual funds, they can efficiently align money
flow with managerial talent. Stein (2005) further argues that when open-ending is the only
creditable signal of managerial ability, it becomes the dominant structure in the mutual fund
industry despite its apparent inefficiency in allowing skilled managers to take advantage of
arbitrage opportunities (e.g., Shleifer and Vishny (1997)). Combining the two literatures,
one is tempted to conclude that the OEF structure is costly in meeting liquidity needs while
it might be necessary for rewarding superior abilities. We add to the debate by showing that,
even in the absence of ability, the OEF structure has its merits relative to the ETF structure.
On the practical front, our analysis suggests that the rush in the industry to develop active
ETFs – in addition to presenting the technical difficulty of revealing the portfolio holding
for in-kind redemption while maintaining anonymity – may be ill motivated.
Finally, we contribute to the literature on Closed-Ended Mutual Funds. While the earlier
literature largely relates the discount/premium of CEFs to irrational behavior of investors
and the cost of arbitrage (see, for example, Lee, Shleifer, and Thaler (1991) and Pontiff
(1996)), recently more effort has been devoted to explaining the discount/premium using
rational models with managerial ability and liquidity demands (see, for example, Chay and
Trzcinka (1999), Berk and Stanton (2007) and Cherkes, Sagi, and Stanton (2007)). Viewing
ETFs as closed-ended index funds, we establish a benchmark at which neither managerial
ability nor a discount/premium exists. Our framework, which captures the main difference
between the OEF and ETF structures, can provide a starting point for understanding the
difference between the ETF and CEF structures and in turn shed light on the phenomenal
growth of ETFs and the stagnation of CEFs (which are active ETFs except for the in-kind
redemption feature).
The remainder of the paper proceeds as follows. Section 2 describes the ETF industry
and how it has evolved. Section 3 sets up the model and Section 4 describes the equilibrium.
Section 5 discusses the properties Finally, Section 6 concludes.
2 The growth of the ETF industry
2.1 Differences between ETFs and OEFs
ETFs are closed-ended investment companies that can be traded at any time throughout
the course of the day on the secondary market. ETFs try to replicate stock market indexes.
Contrary to regular index OEFs, ETFs are required to hold all the stocks that comprise the
6
index they track.
The first ETF was introduced on the Toronto Stock Exchange in 1990. The first U.S.
ETF was introduced in 1993 by State Street Global Advisors – SPDR (Standard & Poor’s
Depository Receipts, or “Spiders”), which tracks the S&P 500 index – and it is still the
largest ETF by market cap. Most stock exchanges around the world now offer ETFs.
ETFs are closed-ended and not open-ended mutual funds and hence have two main
differences from regular open-ended mutual funds. First, they are traded on the secondary
market and therefore can be traded during the day and not only once a day. Second, if
there is an inflow of money to the fund, the fund does not directly create new shares.
However, contrary to standard closed-ended mutual funds, new shares can be created (or
redeemed). The creation and redemption of shares in ETFs is also different from that of
OEFs. Rather than the fund manager dealing directly with shareholders, certain parties, such
as institutional investors, who have entered into a contract with the fund (called Authorized
Participants (APs)) will create a basket of shares replicating the index that the fund tracks,
and deliver them to the fund in exchange for ETF shares. A basket, or creation unit, consists
of anywhere from 10,000 ETF shares to 600,000 ETF shares. ETF shares are then sold and
resold freely among investors on the open market. If an investor accumulates a sufficient
amount of ETF shares, the investor can exchange one full creation unit of ETF shares
for a basket of the underlying shares of stock. The ETF creation unit is then redeemed,
and the underlying stocks are delivered out of the fund. One of the advantages of this
creation/redemption process for the fund investors is that institutional investors cover the
dealing costs in purchasing the required shares to make up the portfolio. This mechanism
allows institutional investors to take advantage of arbitrage opportunities when the price of
the ETF deviates from its Net Asset Value (NAV). It is also the main distinction between an
ETF and a regular closed-ended mutual fund, which results in a very small discount/premium
between the price and the NAV. Thus, none of the regular discount/premium characteristics
discussed in the literature with regard to closed-ended mutual funds apply in any substantial
way to ETFs.
ETFs and OEFs also have different tax advantages. Whenever an OEF realizes a capital
gain that is not balanced by a realized loss, the mutual fund must distribute the capital
gains to shareholders by the end of the quarter. This can happen when stocks are added to
and removed from the index, or when a large number of shares are redeemed. These gains
7
are taxable to all shareholders. In contrast, ETFs are not redeemed by holders (instead,
holders simply sell their ETFs on the stock market, as they would a stock), so investors
generally only realize capital gains when they sell their own shares. However, there are
some potential taxation drawbacks to ETFs. First, ETFs have to hold the exact mirror
image of an index, while OEFs do not. Hence, around changes in the composition of an
index, the ETF may have to sell existing stocks in order to re-balance its holdings. OEFs
do not have to hold the mirror image of the index, and hence have more flexibility around
index changes. Second, ETFs often trade their shares more rapidly to maintain a high cost
basis of their underlying shares, which can result in ETF dividends failing to be classified
as qualified dividends since the underlying shares don’t satisfy IRS requirements. This can
be a substantial drawback since one’s ordinary tax rate may be significantly higher than the
15% tax charged on qualified dividends.
Note that ETFs are structured either as a mutual fund or as a Unit Investment Trust
(UIT). A UIT is a U.S. investment company offering a fixed portfolio of securities that have
a finite life. UITs are assembled by a sponsor and sold through brokers to investors.
2.2 Data
We use several different data sets in generating our sample:
First, we use CRSP’s Survivor-Bias Free U.S. Mutual Funds Database, from which we
extract a list of the index mutual funds (OEFs). We define a fund to be an index mutual
fund if the word index (or any derivative of it such as Ind, Idx, and so on) shows up in its
name. Whenever in doubt, we check the prospectus of the fund using EDGAR, the SEC
Filings and Forms Database, in order to verify that indeed it is a pure index fund. This
step of verification from EDGAR is necessary, as some mutual funds can track an index
closely (and have the word index in their name) and yet not be an index fund. For example,
self described “enhanced index” funds essentially track an index but knowingly change some
weights of stocks they believe will over- or under-perform the index, and hence “enhance”
the index. If the prospectus is not clear that the fund performs only the task of tracking an
index, we err on the side of caution and drop the fund from our sample. We then merge the
different share classes of each fund into one entity. As has been recognized in the literature,
the same mutual fund can have several almost identical share classes. The different share
classes usually differ either in their fee structure (i.e., front-end load, back-end load, no load,
8
and so on) or in their distribution channel (i.e., investor class, institutional class, and so on).
We merge the different share classes into one OEF entity by value weighting the different
classes based on their Total Net Asset (TNA). The resulting sample is comprised of 296
OEFs over the time span of 1992-2006. These funds are managed by 131 different families.
Interestingly, these 296 funds track only 63 different indexes.
Second, we use Bloomberg in order to generate a list of ETFs. We collect the ticker
and CUSIP of each ETF, we also find out what fund family is sponsoring it. In order to
verify that we have captured the full span of ETFs, we check the websites of the respective
sponsoring families and ETF-dedicated websites (such as ETFConnect.com). The resulting
sample is comprised of 320 ETFs over the time span of 1992-2006. We restrict both our
ETF and OEF samples to start in 1992 due to the fact that there were no ETFs before then.
This allows us to capture the entire current growth of ETFs from their creation to 2006. We
exclude any ETF that existed less than one year, as we would not have enough information
about it. These 320 ETFs are sponsored by 23 separate families, and they track 268 different
indexes. Thus, the final sample consists of 296 OEFs and 320 ETFs over 1992-2006, tracking
289 different indexes.
Third, we collect information about the ETFs and OEFs such as assets under manage-
ment, number of shares outstanding, creation and deleting date, from CRSP and CRSP
Mutual Fund merging by ticker and year.
Last, in order to analyze the indexes tracked by these funds, we need their historical
compositions. These compositions are not available for all the indexes. Hence, we employed
a proxy for their composition. We use Thomson Financials ’s CDA/Spectrum Mutual Funds
Holding and CDA/Spectrum Institutional (13f) Holdings. For each index we find all the
OEFs and ETFs that track it, since they have to report their holdings on a quarterly basis.
Based on the holdings we calculate the variables relating to volatility and composition of the
index.
2.3 Trends in the ETF industry
As we discuss in the introduction, while ETFs were introduced in the U.S. only in 1993, they
have since grown at an exponential rate from one ETF to more than 300 in 2006.
Figure 1(a) shows the significant increase in money invested in ETFs between the early
1990s and the present. As one can see, the growth in ETFs was clearly slow between 1993
9
(a) Dollar Amount Invested (b) Number of Funds
0
100
200
300
400
500
600
700
1992 1994 1996 1998 2000 2002 2004 2006
200
250
300
350
0
50
100
150
1992 1994 1996 1998 2000 2002 2004 2006
(c) Number of Indexes Tracked (d) Average Total Net Assets
200
250
300
0
50
100
150
1992 1994 1996 1998 2000 2002 2004 2006
2000
2500
3000
3500
0
500
1000
1500
1992 1994 1996 1998 2000 2002 2004 2006
Figure 1: Summary statistics of the ETF and the OEF industries. In all panels,the dotted and the solid lines refer to OEFs and ETFs, respectively. Panel (a) reports thedollar amount invested (in billion dollars), panel (b) reports the number of funds, panel (c)reports the number of indexes tracked, and panel (d) reports the average total net assets ofthe funds (in million dollars).
10
and 2000. This is for two main reasons. First, ETFs were a new investment vehicle, and
as such were not yet very familiar to investors. Second, there were few ETFs available,
making it a relatively exotic investment vehicle. Figure 1(b) depicts the relative number
of ETFs available compared to OEFs over time. By year 2000, there were only about 20
ETFs available covering the major indexes, while there were more than 150 OEFs available.
A direct comparison of the number of available ETFs to the number of available OEFs,
however, is slightly misleading. In the early years each new ETF tracked a different index;
even now, there are few ETFs that track exactly the same index. OEFs, in contrast, work
somewhat differently. Mutual funds depend on their respective distribution channels, and
not all mutual funds are available to all investors. For example, an investor with a Vanguard
account can only invest in Vanguard mutual funds. Investors in retirement accounts follow
even stricter rules and have limited offerings that depend on the company administering the
account. The result is that in the past 40 years each mutual fund company (or at least the
larger ones with their own distribution channels) has come to offer funds that are very similar
to those of other mutual fund families. In the case of index mutual funds many families offer
essentially identical funds that track the same index. For example, there are more than a
dozen OEFs that track the S&P 500, while not long ago there was only one ETF. Even now
there are only two ETFs tracking this index. Thus, while there are more OEFs than ETFs
until very recently, OEFs are tracking fewer indexes.
Figure 1(c) shows the large increase in the number of different indexes tracked by ETFs
in the past seven years. Comparing Figure 1(c) to 1(a), we can see that part of the success
of ETFs and the apparent market share they seem to have captured from OEFs is correlated
with the huge increase in the number of indexes that are being tracked. This connection
is related to our theoretical argument that the characteristics of the underlying indexes
tracked may be related to the increased interest in these indexes. As Figure 1(a) shows,
in the past 10 years, while ETFs have grown at a faster pace than OEFs, increasing their
relative importance in the marketplace, they have also offered new indexes as investment
vehicles at an exponential rate. Together, these figures imply that, indeed, there could be
more to the rise in ETFs than the mere fact that they offer an advantage to small investors
compared to traditional index OEFs.
Though this impressive growth of ETFs can be partially attributed to the diversity of
offerings in terms of the number of indexes covered, Figure 1(d) indicates that the average
11
ETF has comparable size to the average OEF. Admittedly, since there are many OEFs
tracking each index and the decision of small investors to invest in them depends on their
respective distribution channels, this is far from conclusive. It does indicate, however, that
the growth process of this relatively new industry is a compilation of several simultaneous
trends: The intrinsic growth of existing ETFs, the introduction of new ETFs covering existing
indexes, and the offering of ETFs covering new indexes that are not offered by traditional
OEFs. This evidence highlights the importance of the question at the root of this paper,
namely, whether the apparent increase in the size of the ETF industry is going to eventually
alter the landscape of the delegated portfolio management industry and, in particular, of the
existing mutual fund structure.
2.4 The Growth of the ETF Industry
Since the universe of ETFs and OEFs has been quickly changing, and since the growth rate
has been so substantial and different between OEFs and ETFs, we describe the time trends
in the data in further detail in Table 1. Though introduced in the 1970s, by the 1990s
index OEFs were still not very popular. In 1991 there were only 34 OEFs offered by 19
families managing about 11 billion dollars. ETFs entered the market in 1993 when there
was a large inflow in index OEFs. As mentioned in Section 2, though introduced in 1993,
ETFs took several years to get exposure. In 1996 assets under management in OEFs had
increased to more than 91 billion dollars and ETFs had 3 billion under management. This
fact is thoroughly discussed in Svetina and Wahal (2008) and ICI (2008). This relationship,
however, is reversed when looking at the number of indexes tracked by ETFs and OEFs.
The 296 OEFs track only 63 different indexes, while the 320 ETFs track 268 different ETFs.
Similar to the argument about the level of competition in the two markets made by Hortacsu
and Syverson (2004), we can see that ETFs have generated large market differentiation by
tracking different indexes, while OEFs, protected by their distribution networks track a
substantially smaller and similar list of indexes.
By 2001, the trend of growth had flipped between OEFs and ETFs. In 2001 ETFs were
already growing at a rate three times faster than OEFs. The average growth rate in 2001 of
index OEFs was only 7% and of ETFs was 389%. Several interesting trends can be observed
in the data. Both the number of families offering OEFs and offering ETFs increased between
1996 and 2006, from 40 to 72 for the OEFs and from 2 to 23 for the ETFs. This highlights
12
two interesting facts. First, the rise in the interest in indexing in the late 1990s brought an
increase in offering of index funds by many families. Second, relatively few families offered
ETFs even though they were growing at a much faster pace than OEFs, and even though
offering one or the other is not technically very different. Vanguard is an example of a family
that essentially offers mainly index OEFs and yet refused for many years (under the influence
of John Bogle) to also offer ETFs. These facts explain the much higher growth in assets
under management per family in ETFs than in OEFs. As we mentioned earlier, OEFs are
not always available to all investors depending on the family’s distribution channel, making it
more difficult at times to grow. By being available to all investors with a brokerage account,
ETFs have enjoyed an advantage in growth that is fairly apparent in the data.
As shown in Table 1, an important aspect is the number of indexes tracked offered by
each sponsoring family. This number has steadily been increasing for ETF sponsors, from
9 per family in 1996 to more than 13 in 2001. During the same time period OEF sponsors
have only marginally increased the number of different indexes covered from 2 in 1996 to 2.3
in 2001. This highlights that there might be two different forms of growth not shared by the
two organizational forms that is captured by the number of covered indexes offered.
2.5 Indexes tracked
As we show in Table 1 a large part of the growth of the ETF industry is through the offering
of many different ETFs by each family, each one tracking a different index. One of the
main motivation of this paper is to better understand this growth and the reasons behind
it. In order to investigate this growth, we calculate characteristics of these indexes in order
to assess the nature of the ETF growth. In Table 2 we describe the characteristics of the
different indexes. The diversity across different indexes is quite large, they vary from tracking
more than two thousand stocks to as few as five stocks. The market capitalization of these
indexes is also quite diverse, ranging from small “niche” indexes tracking a segment of 35
billion dollars to market-wide indexes tracking more than 1 trillion dollars. We look at four
main characteristics of these indexes: the size of the index, the number of index changes, the
liquidity of the underlying stocks, and the concentration of the industry tracked. As expected,
indexes vary significantly on all these dimensions. One of the larger costs associated with
tracking an index is the cost of re-balancing. The average number of stocks entering/exiting
the index is 18 stocks a quarter, with an average market value of 119 billion dollars, so these
13
costs can be very high. The tracking error generated by re-balancing the index is higher for
less liquid and smaller stocks. There is also significant variation in the required re-balancing
across indexes – some indexes require almost no re-balancing while others require selling and
buying more than 60 stocks in a quarter.
We use Amihud (2002)’s liquidity measure. We calculate it by downloading the measure
for all stocks from Joel Hasbrouck’s website (which is described in detail in Hasbrouck (2006))
and then averaging the underlying liquidity measures of the stocks comprising the index, both
as a value-weighted average and as an equal-weighted average. Again, the difference across
indexes is fairly substantial. Some indexes have relatively high liquidity stocks (the S&P
500, for example) and some have fairly low liquidity (some industry specific indexes, for
example). Since the liquidity of the underlying index directly affects the tracking error of
the OEF, we know from our theoretical argument that the best vehicle (OEFs or ETFs)
tracking a particular index is dependent on the index liquidity. We calculate the level of
industry concentration using two different measures: we use a Herfindahl-Hirschman Index
and the industry concentration measure developed in Kacperczyk, Sialm, and Zheng (2007).
As we can see in Table 2, there is variation on this level too.
3 Model
We construct a parsimonious model that captures the main difference between OEFs and
ETFs: Orders are cleared in the ETF market directly and traders bear their own trading
costs, whereas orders are pooled in the OEF before being submitted to the underlying stock
market and all investors in the OEF (whether or not they trade) equally share any resulting
trading costs. We model agents’ trading needs by introducing heterogeneity in their risk
exposures, which motivates them to trade for risk sharing.
3.1 Securities market
There are three dates, t = 0, 1, and 2. A risky stock S is traded in a competitive market.
The stock yields a final payoff of V at time 2, which is normally distributed,
V ∼ N(V , σ2
v). (1)
The per capita supply of the stock is θ. In addition, there is a short-term riskless security,
which yields a constant interest rate of zero.
14
In addition to trading the stock directly, investors can participate in either an OEF
(denoted by F ) or an ETF (denoted by E). Our main objective is to compare the effectiveness
of the OEF and the ETF structure in meeting liquidity needs for individual investors. To
simplify analysis, we assume that the ETF is fully integrated with the underlying stock
market, that is, the stock is identical to the ETF and is omitted from the setting. We
use the ETF and the underlying stock interchangeably in our discussion depending on the
context.9
We model the OEF structure in the following reduced-form way: At time 0, OEF investors
give their stock endowments to the fund in exchange for OEF shares. Each OEF share is
normalized to be equal to one share of the stock.10 At time 1, any OEF investor can redeem or
purchase any number of shares at a pre-specified price of P F1 per share, which is independent
of the market condition at time 1. The fund can borrow or invest at the risk free rate to
accommodate the fund flow. That is, if in aggregate OEF investors demand extra shares at
time 1, the fund issues new shares in exchange for cash payments from its investors and then
parks the money in the risk free asset; if instead the aggregate demand is a sell order, the
fund borrows money and pays out to the departing investors at the price P F1 . Shortly after
that, at time 1+, the fund passes the demand to the underlying stock market. That is, they
buy or sell stock shares at the market price P E1 to ensure that each remaining OEF share
has the same risk exposure as one share of stock. In this sense, the OEF is extremely passive
– it does not optimally manage the cash position to account for potential price impacts. To
the extent that P E1 6= P F
1 , the fund makes or loses money on these transactions. The gains
or losses are shared evenly by all remaining shares in the OEF, leading to a terminal payoff
different from that of the stock. We term the difference the “tracking error” of the OEF.
As will become obvious later, the price difference between P F1 and P E
1 is predictable at
time 1. To rule out arbitrage, we assume that once investors choose to invest in an OEF,
they cannot access the ETF or the stock market directly.
9In reality, an ETF is a portfolio of stocks. It may have a tracking error relative to the underlyingstock index and may have different liquidity (can be either higher or lower) from its component stocks.The possibility of arbitrage between the ETF and its underlying stocks guarantees a small tracking errorand partially alleviates this concern. Moreover, the main objective of this paper is to compare the relativeefficiency of OEFs and ETFs rather than to compare their absolute efficiency relative to the underlyingstocks. We leave the detailed comparison between the ETF and its underlying stocks to future research.
10In practice, to start investing in an OEF, investors pay cash and the fund purchases stocks. We assumeOEFs are formed by accepting stocks directly from their investors to eliminate the impact of OEF investingon time 0 stock prices. This price impact complicates the analysis and does not affect our main message.
15
3.2 Agents
There is a continuum of investors in the economy with a total population weight of 1 +µ, of
which µ fraction are stock investors and the remaining (normalized to 1) are index investors
choosing between the OEF and the ETF. The focus of the paper is the decision of indexers
between the two indexing vehicles. Stock investors are introduced so that there is always a
viable stock market in which the OEF can transact (at time 1+) to implement the demand
from its investors, even when the population of ETF investors goes to 0.
Each investor is endowed with θ shares of the stock at time 0. Investor i also receives a
non-traded payoff Ni at the terminal date 2, which is given by
Ni = Yi (V − V ), Yi = Y + ǫi, (2)
where V is the stock payoff in (1), and Y and ǫi are mutually independent, normally dis-
tributed random variables with a mean of zero and a volatility of σY and σi, respectively.11
We can interpret this non-traded income as a liquidity shock that is correlated with the
stock. In particular, it is equivalent to an endowment shock of Yi shares of the stock. Given
the correlation between the liquidity shock and the stock payoff, investors want to adjust
their stock positions in order to hedge this risk, giving rise to their trading needs.12
All investors are subject to the same Y shock. Hence, it captures the aggregate liquidity
shock. The component ǫi is independent across all investors and defines their individual
shock. Some investors face more individual liquidity shocks than others, so we assume
σi = si σǫ, si ∼ Unif [0, 1], (3)
where σǫ is the maximum individual liquidity shock, and si is the magnitude of liquidity needs
for individual i, which is uniformly distributed over the range [0, 1]. Since ǫi is independent
with bounded variance and each individual investor has zero population weight, Law of Large
Number holds. Thus, for any subset I of investors, individual liquidity shocks always cancel
out and the total liquidity demand depends only on the aggregate shock and is proportional
11We assume that the non-traded payoff Ni is perfectly correlated with the stock payoff V . As long as Ni
and V are correlated, the qualitative nature of our results is independent of the sign and the magnitude ofthe correlation.
12Heterogeneity in endowment is merely a device to introduce the need of trading for risk-sharing purposesas in Diamond and Verrecchia (1981) and Huang and Wang (2008). Other forms of heterogeneity can alsogenerate trading needs, such as difference in preferences or beliefs. Our modeling choice is mainly motivatedby tractability.
16
to the population weight µI ,∫
i∈I
ǫi = 0, and
∫
i∈I
Ni = µI Y (V − V ). (4)
All agents have identical preference, which can be described by an expected utility func-
tion over the terminal wealth. For tractability, we assume that the agent exhibits mean-
variance preference. In particular, agent i has the following utility function:
E[W i2] −
γ
2Var[W i
2], (5)
where γ is the risk aversion and W i2 is the agent’s terminal wealth.
3.3 Time line
We now describe in detail the timing of events and actions. At time 0, indexers decide
whether to invest via the ETF or the OEF. The market equilibrium determines the fractions
η and 1 − η of investors who choose to invest in the ETF and the OEF, respectively. ETF
investors do nothing at time 0, and OEF investors exchange all their stock shares for an
equal number of OEF shares.
0 1 1+ 2
Shocks Yi = Y + ǫi V , Ni
OEF investors θ, 1 − η θFi1, XF ; XF
ETF investors θ, η θEi1
Stock investors θ, µ θSi1
Equilibrium η
PE1
∆PF1
PE2 = V
PF2 = V −∆
Figure 2: The time line of the economy.
At time 1, investors in the OEF learn the aggregate liquidity shock Y and their individual
risk exposure ǫi. They take as given the price P F1 and choose their optimal holding θF
i1.
Investor i redeems (θ − θFi1) shares from the OEF and receives a payment of P F
1 (θ − θFi1) (or
pays cash to purchase shares if (θ − θFi1) < 0). Aggregating over all OEF investors, the OEF
experiences a total redemption of
XF ≡
∫
i∈OEF
(θ − θFi1) = (1 − η) θ −
∫
i∈OEF
θFi1 (6)
17
shares (or creation if XF <0). Hence, the total OEF shares outstanding is∫
iθF
i1. The assets
of the OEF consist of (1 − η)θ shares of the stock and −XF P F1 dollars of cash.
Note that the supply of OEF shares is not fixed at time 1 and is a function of the pre-
determined price P F1 , since the fund can create or redeem any number of shares based on
investor demand. In this sense, P F1 is not an equilibrium price that clears the market, but
rather a contractual price that all OEF investors agree upon as a part of the OEF contract.
In the model, we impose an additional restriction that the aggregate trading demand from
OEF investors (XF ) is zero when the aggregate liquidity shock is Y = 0 to pin down the
price.
At time 1+, the OEF passes the aggregate demand from its investors through to the
underlying asset market by buying or selling stocks. It sells exactly XF shares of the stock,
independent of the transaction price P E1 , to make sure that each of the remaining OEF shares
has the same risk exposure as a share of the stock. After the transaction, the remaining assets
in the OEF include (1 − η) θ − XF =∫
i∈OEFθF
i1 shares of the stock and −XF P F1 + XFP E
1
dollars of cash, to be shared equally among all the remaining OEF shares. Thus, each OEF
share is equivalent to a share of the stock plus a loss of −∆, defined as
∆ = XF (P F1 −P E
1 )/(
∫
i∈OEF
θFi1
)
. (7)
At time 1+, in the ETF market, all stock and ETF investors participate both to unload
their own shock Yi and to accommodate the above demand XF from the OEF. To simplify
notation, we use the same time index 1 for all variables realized at 1+. In particular, we use
P E1 to denote the equilibrium ETF/stock price and θE
i1 to denote the holdings of an ETF
investor after trading at time 1+. There is little room for confusion, since we assume that
OEF investors are not allowed to trade in the ETF. This assumption is needed to rule out
arbitrage opportunities given the price difference between P F1 and P E
1 .
At time 2, investors liquidate their OEF or ETF holdings at prices P F2 and P E
2 , respec-
tively. Obviously, P E2 = V . The OEF has a per share loss of ∆ dollars in addition to its
stock holding. Hence, P F2 = V −∆. We refer to ∆ as the tracking error of the OEF relative
to the ETF. Including their non-traded labor income in (2), the terminal wealth for OEF
and ETF investors are, respectively,
W Ei2 = θ P E
1 + θEi1(V − P E
1 ) + Yi (V − V ) (8a)
W Fi2 = θ P F
1 + θFi1(V − ∆ − P F
1 ) + Yi (V − V ). (8b)
18
3.4 Discussions of the model
In this subsection, we provide additional discussions and motivations about several important
features of the model.
First, since we equate ETFs with the underlying stocks, all our discussion of the benefit
and the cost of the OEF structure should be interpreted as a relative statement in relation
to the ETF structure. For example, the diversification benefit of indexing is common to
both structures and does not affect the comparison. Also, even though historically ETFs
are introduced after OEFs and the natural question to ask is whether ETFs add value,
technically our model asks the opposite question of whether OEFs add value above and
beyond what ETFs can offer.
Second, while our specification of the OEF and the ETF trading mechanisms captures
their key difference in terms of flow-induced trading costs, we deviate from the true trading
mechanisms for tractability. In practice, for most OEFs, investors submit their purchase or
redemption orders throughout the day to meet their liquidity needs Y . All their orders are
pooled and executed at the end of day Net Asset Value (NAV), calculated using the closing
prices of their stock holdings at 4pm. The OEF observes preliminary signals about fund
flows and may transact in the underlying stock market throughout the day in anticipation of
these demands. However, since on average OEFs do not observe the full flows until several
hours after the market closure, they need to execute the remaining orders during the next
trading day (or even several days afterwards if the demand is large). On the other hand,
ETF investors trade sequentially at different prices depending on their time of entry into the
market and bear their own transaction costs.
The above description highlights two sources of flow-induced trading costs for the average
OEF investors relative to ETF investors. First, given an order flow, the overall transaction
costs can differ between the OEF and the ETF. The average transaction price might be
worse for the OEF because of its delay in observing the full order flow or because of the
(perceived) superior investment ability of the average ETF investors. Alternatively, the
average price might be better for the OEF if the OEF manager can add value by forecasting
the overall liquidity demand or by implementing the orders more efficiently than the average
ETF investors throughout the day. Second, even if the OEF and the ETF receive the same
average transaction prices, they may allocate the overall costs differently among investors.
In particular, the OEF may allocate part of the costs to investors not incurring any trades.
19
The reason is that all orders entered before 4pm receive the end-of-day NAV, which does
not incorporate the future price impact that the OEF may experience when it implements
the remaining demand in the stock market the following day. From the literature on costly
mutual fund flows (e.g., Edelen (1999) and Greene and Hodges (2002)), we know that the
combined effects of these two sources of flow-induced trading costs are large for OEFs.
While it is hard to separate the two types of flow-induced trading costs, in the model we
rule out the differential overall transaction costs by assuming a unique price P E1 at which both
the ETF investors and the OEF transact. Thus, we do not compare the absolute efficiency
of the OEF and the ETF in implementing their liquidity trades. Rather, our model focuses
on the relative efficiency of the two structures given their differential allocation of the overall
transaction costs: The OEF structure provides partial insurance against the liquidity shock
by setting a price P F1 that is less dependent on Y ; the ETF structure provides no insurance
by trading at P E1 directly. Although in practice the end-of-day NAV P F
1 may depend on the
part of Y that is revealed during the day, it is reasonable that P F1 does not fully incorporate
the price impact of Y , since some OEF flows are not observable even to the OEF itself till
the next day. Thus, P F1 is less dependent on Y than P E
1 is and the OEF structure provides
some liquidity insurance. In the model, we assume a constant P F1 for tractability.
Third, in reality, ETF investors need to pay bid-ask spreads to the stock exchange whereas
the OEF charges zero bid-ask spread by providing internal crossing of individual trades. The
assumption that all ETF investors transact at price P E1 ignores this difference and overstates
the benefits of the ETF. Moreover, this assumption of a batched price P E1 smooths the utility
of all ETF investors across the different transaction prices they may receive throughout the
day, further increasing the estimated benefit of the ETF structure. This assumption is
conservative for our main conclusion that the OEF structure is not dominated.
Fourth, ETF is generally perceived as more tax-efficient due to its ability to pass out low-
tax-basis stocks to authorized investors via the in-kind redemption feature. However, Poterba
and Shoven (2002) find that although in theory ETFs can be more tax efficient, in reality
SPDR ETF performs slightly worse than the Vanguard S&P 500 both in before-tax and
after-tax returns. In general, ETFs become less tax-efficient when they are forced to trade
frequently to accommodate index changes or to meet redemption needs.13 In addition, the
13On Dec. 10, 2008, the Rydex Inverse 2X S&P Select Sector Energy ETF paid out 87% of its total assetsunder management as short-term capital gains, highlighting the extreme tax inefficiency of some ETFs. GusSauter, CIO of Vanguard, also encourages investors not to be oversold on the ETF tax-efficiency claims in his
20
tax-law change of 2003 allows a reduced tax rates (at 15%) for qualified dividends. Utilizing
the in-kind redemption process reduces the holding period and may disqualify dividends for
the lower tax rate.
Fifth, we abstract away some institutional details to highlight the structural difference
between OEFs and ETFs. For example, an OEF can replicate an index with a small number
of stocks while an ETF organized as a unit trust has to hold the whole index. Even for an
ETF that is organized as an open-ended fund and that is not required to hold the whole
index, more trading is needed during index membership changes given its in ability to manage
the index changes through fund flows. Therefore, OEFs have an advantage in transactions
costs during index changes. Moreover, we abstract away other operational costs for OEFs,
like order execution, book keeping, and so on. These costs have a large fixed component and
decrease with the fund size.
4 Equilibrium
We solve the equilibrium backwards in three steps. First, taking the demand from the OEF
as given, we solve the optimization problem of ETF investors to derive the equilibrium price
at time 1+. Second, we solve the trading decision of OEF investors at time 1 and the
equilibrium tracking error. Third, we evaluate the utility of investing via the OEF and the
ETF for investors with different liquidity needs at time 0 and determine the equilibrium
fraction of investors who choose to invest in each.
4.1 Equilibrium at time 1+ in the ETF market
Assume the aggregate demand from OEF investors is XF . The following proposition solves
the optimal decision of ETF investors and the market equilibrium at time 1+.
Proposition 1. Given the population η (and µ) of ETF (and stock) investors, the demand
XF from OEF investors, and the aggregate liquidity shock Y , the equilibrium ETF price at
time 1+ is
P E1 = V − γσ2
v θ − γσ2
v
(
Y +XF
η + µ
)
, (9)
article, “Index chief weighs in on 30 years of indexing, touted new ‘indexes’,” at https://institutional.vanguard.com/VGApp/iip/site/institutional/researchcommentary/article?File=NewsIndexChief.
21
and the optimal holding of an ETF (and stock) investor i is
θEi1 = θS
i1 =V − P E
1
γσ2v
− Y − ǫi (10a)
= θ +XF
η + µ− ǫi. (10b)
By construction, once the ETF investors decide to invest via the ETF, they are identical to
stock investors in the model. From equation (10a), the desired holding of investor i depends
on the risk-return tradeoff of the stock (the first term) plus a hedging term against the
liquidity shock Yi = Y + ǫi. Since Y is the aggregate shock, it affects everyone’s desired
holding and hence the equilibrium price in (9). In equilibrium, the price P E1 fully adjusts for
the impact of Y and investors no longer choose to unload Y , as (10b) indicates. In summary,
the aggregate risk Y affects only the price but not the trading decisions in equilibrium. All
ETF and stock investors equally share the aggregate risk Y and they unload only their
idiosyncratic risk exposure, ǫi, in the market. The excess demand, XF , from the OEF
investors, is an aggregate risk for the ETF and stock investors (with total population weight
η + µ). They equally share this risk and the equilibrium price incorporates this risk as well.
4.2 Equilibrium at time 1 in the OEF market
We now solve the OEF equilibrium at time 1 in two steps. First, taking as given the
functional forms of both the price P F1 and the tracking error ∆, OEF investors choose their
optimal holding conditional on their liquidity shock. Second, we aggregate the demand from
OEF investors and solve for equilibrium prices and the tracking error.
As stated earlier, we impose the additional restriction that the aggregate trading demand
from OEF investors (XF ) is zero when the aggregate liquidity shock is Y = 0 to pin down
the price. The following proposition derives the equilibrium price P F1 and individual share
holdings based on this restriction.
Proposition 2. Assume the OEF price is P F2 = V − ∆ at time 2, where ∆ = ∆0 + ∆1Y +
∆2Y2 is the tracking error, and that the aggregate trading demand from OEF investors (XF )
is zero when the aggregate shock is Y = 0. Then the OEF price at time 1 is
P F1 = V − γσ2
v θ − ∆0, (11)
22
and the optimal holding of an OEF investor i is
θFi1 =
V − ∆ − P F1
γσ2v
− Y − ǫi, (12a)
= θ −∆1Y + ∆2Y
2
γσ2v
− Y − ǫi. (12b)
The holding in (12a) is similar to that of (10a), except that the tracking error ∆ lowers the
expected future payoff of the OEF and reduces the demand for the stock. Similar to the
ETF case, ignoring the functional form of the price, investors would like to hedge both the
aggregate shocks Y and the idiosyncratic shock ǫi. In contrast to the ETF case, however,
the equilibrium price P F1 does not fully account for the impact of Y . In particular, we model
a special case in which P F1 is independent of Y . As a result, OEF investors optimally unload
their aggregate risk exposure Y in equilibrium. This is in direct contrast to the case of ETF
investors in Proposition 1, who hedge only the idiosyncratic risk ǫi. The difference between
the trading strategies for the OEF and the ETF investors reflects a “moral hazard” effect.
Since OEF investors are guaranteed a price P F1 independent of Y , they have the incentive
to unload their aggregate risk exposure Y even though in equilibrium it is more efficient for
everyone to share the aggregate risk.
We now connect the OEF and the ETF market through the definitions of OEF demand
XF and the tracking error ∆ in (6) and (7), respectively. The tracking error is a complicated
function of Y . We use Taylor expansion to expand it around Y . As long as Y is small relative
to the total supply of the stock, we can drop higher-order terms. In particular, we drop terms
higher than the second order and match coefficients of the Y terms to derive the equilibrium
in the following Proposition.
Proposition 3. When Y is small (relative to θ),
(i) at time 2, the equilibrium OEF price is P F2 = V−∆, where ∆ is the tracking error and
∆ = ∆2Y2, ∆2 ≡ (1 + η)γσ2
v/θ, η ≡1 − η
η + µ(13)
(ii) at time 1, the equilibrium prices of the OEF and the ETF are, respectively,
P F1 = V − γσ2
v θ (14a)
P E1 = V − γσ2
v θ − γσ2
v(1 + η)Y − η∆2Y2 (14b)
23
(iii) at time 1, the equilibrium holdings of the OEF and the ETF investors are, respectively,
θFi1 = θ − ǫi − Y − (1 + η)Y 2/θ (15a)
θEi1 = θ − ǫi + ηY + η(1 + η)Y 2/θ, (15b)
and the aggregate demand from OEF investors is XF = (η + µ)(x1Y + x2Y2), where
x1 = η and x2 = η(1+η)/θ.
Note that the population weight of OEF investors is 1−η and the population weight of ETF
and stock investors is η+µ, hence, η = (1−η)/(η+µ) in (13) defines the relative population
weight of the OEF investors. From (15), OEF investors unload all of their aggregate risk
exposure (Y ) to the ETF market, while ETF investors accommodate the demand. This extra
demand causes an additional impact on the price P E1 and leads to the OEF tracking error
∆. Although collectively costly, individual OEF investors have little incentive to internalize
this cost given their guaranteed price P F1 .
4.3 Value functions of the OEF and the ETF investors at time 0
We calculate the value functions of the OEF and the ETF investors in two steps, taking as
given the fraction of investors that choose to invest in each. First, we consider the partial
equilibrium effect when both the tracking error ∆ of the OEF and the order flow XF faced by
ETF investors are taken as exogenously given. Second, we consider the general equilibrium
effect by incorporating the equilibrium ∆ and XF from Proposition 3. These two steps allow
us to separate the effect of individual optimization from the equilibrium price impact.
The following lemma derives the individual value functions of OEF and ETF investors
taking ∆ and XF as given. To simplify the expressions, we use the knowledge from Propo-
sition 3 that ∆0 = ∆1 = x0=0.
Lemma 1. Assume the tracking error of the OEF is ∆ = ∆2Y2 and the ETF market has
a demand shock of XF = (η + µ)(x1Y + x2Y2). When σy is small (relative to θ), for an
investor anticipating an idiosyncratic liquidity shock of the size σi = siσǫ, the value function
of investing in the OEF and the ETF is, respectively,
JFi0 = θV −
1
2γ
(
1+s2
i kǫ+ky
)
kθ − (1−s2
i kǫ) θσ2
y∆2 + O(σ4
y), (16a)
JEi0 = θV −
1
2γ
(
1+s2
i kǫ
)
(ky+kθ)+ky
2γ(1−kθ)x
2
1−ky
2γ(2x1+x2
1+2x2θ)s2
i kǫ+O(σ4
y), (16b)
where O(σ4y) are terms of the order σ4
y or higher, ky ≡ γ2σ2vσ
2y , kǫ ≡ γ2σ2
vσ2ǫ , and kθ ≡ γ2σ2
v θ2.
24
The value function exhibits several interesting features. First, if ∆ = 0 and XF = 0, the
gain of investing via the OEF (relative to investing via the ETF, JFi0 −JE
i0) increases with the
size of the individual liquidity shock si. The reason is that the liquidity insurance provided
by the OEF structure is more beneficial for investors with high liquidity needs.
Second, the tracking error (∆2) reduces the utility for OEF investors, especially for those
with smaller idiosyncratic liquidity needs (si). Although this result appears similar to the
common perception that investors with low liquidity needs are subsidizing others with higher
liquidity needs, the intuition is different. From (4), all idiosyncratic liquidity needs cancel
out at the fund level. Hence, the only liquidity need that is costly to the fund is the aggregate
liquidity shock Y . It is important to note that since all investors have equal exposure to
the aggregate liquidity shock, they contribute equally to the aggregate trading need and the
resulting tracking error of the fund, whether they have high or low idiosyncratic liquidity
needs.
The same price impact and tracking error, however, can affect their utility differently.
Surprisingly, it is more costly for investors with low idiosyncratic liquidity needs, as indicated
by the third term in (16a), −(1−s2i kǫ) θσ2
y∆2. The intuition is that the price impact and
tracking error depend only on the aggregate shock, which is almost perfectly correlated with
the trading needs of investors with low idiosyncratic liquidity needs. Hence, almost all their
trades occur at a bad time – they tend to sell when the aggregate liquidity shock pushes
down the price and buy when the aggregate liquidity shock drives up the price. On the other
hand, for investors with high idiosyncratic trading needs, their trading is less correlated with
the aggregate shock, and they sometimes lose and sometimes gain from the price discrepancy
between the OEF and the ETF and the resulting track error.
Third, the demand from OEF investors have dual impacts on the ETF investors. On the
one hand, ETF investors make markets for the OEF investors and earn a profit by doing
so. This is reflected in the third term in (16b), ky
2γ(1−kθ)x
21. On the other hand, the extra
demand makes price P E1 more volatile and is costly especially for those with high idiosyncratic
liquidity needs, as indicated by the negative term after that, −ky
2γ(2x1+x2
1+2x2θ)s2i kǫ.
These results suggest that individuals with high idiosyncratic liquidity needs are ill-
suited to provide liquidity to others via the ETF structure because they bear a high cost of
transacting directly at volatile prices and benefit more from the liquidity insurance embedded
in the OEF structure. Hence, there is a natural separation between investors with high and
25
low liquidity needs in their preferred investment vehicle.
In Proposition 4, we incorporate the equilibrium tracking error ∆ and OEF demand XF
from Proposition 3 into the value functions in Lemma 1 to derive the equilibrium utility gain
of investing via the OEF.
Proposition 4. Let η be the fraction of investors who choose to invest via the ETF. When
σy is small (relative to θ), the equilibrium utility gain of investing via the OEF relative to
the ETF for an investor with idiosyncratic liquidity shock si is
GFi (si, η) ≡ JF
i0 − JEi0 =
(1 + η)2
2γ
(
3s2
i kǫ − 1 −1 − η
1 + ηkθ
)
ky + O(σ4
y). (17)
(i) Investor i invests in the OEF if and only if GFi ≥ 0, otherwise, he invests in the ETF;
(ii) GFi increases with individuals’ idiosyncratic liquidity shock, that is, (∂GF
i )/(∂si) > 0.
Proposition 4 suggests that individuals with higher liquidity needs prefer the OEF structure
relative to the ETF. One might question the viability of the OEF structure in equilibrium if
high-liquidity-need investors are their main clientele. Interestingly, since individual liquidity
needs cancel out at the fund level, high-liquidity-need investors do not lead to more volatile
fund flows or higher flow-induced trading costs. As a result, the OEF is not dominated by
the ETF in equilibrium.
4.4 The equilibrium size of the ETF industry at time 0
We solve for the equilibrium size of the ETF industry. The following proposition shows the
existence of equilibrium in which the marginal investors are indifferent between investing via
the OEF or the ETF and derives the equilibrium size of ETFs.
Proposition 5. When σy is small (relative to θ), let η solves the following equation,
η =
0, if GFi (0, 0) > 0;
1, if GFi (1, 1) < 0;
GFi (η, η) = 0, o.w.,
(18)
then all indexing investors with si ≤ η invests via the ETF and the rest (si > η) invests via
the OEF. The equilibrium population of ETF investors is η.
The proof of the proposition is straightforward. When GFi (0, 0) > 0, for any investors with
si > 0, we have GFi (si, 0) ≥ GF
i (0, 0) > 0 since GFi increases in si. Hence, at η = 0 the
26
OEF is preferred by all investors, confirming that η = 0 is an equilibrium. Similarly, when
GFi (1, 1) < 0, we have GF
i (si, 1) ≤ GFi (1, 1) < 0 for any si and ETF is preferred by everyone,
thus, η = 1. The rest of the proof can be found in the Appendix.
While intuitive, the proposition does not give simple conditions on the underlying pa-
rameters for the equilibrium. The following corollary provides a more explicit solution.
Corollary 1. The equilibrium population of ETF investors can be expressed as follows:
η =
0, if µ < (kθ − 1)/(kθ + 1);
1, if kǫ < (1 + kθ)/3;
1
3(1+µ)kǫ
(
kθ +
√
kθ2+3kǫ(1+µ)2+3kǫkθ(µ2−1)
)
, o.w.
(19)
The first case of the corollary indicates that the ETF population drops to 0 for small µ.
This is a situation with very few stock investors to make markets. If the equilibrium fraction
of ETF is small, the price becomes extremely volatile whenever OEF investors unloads
their aggregate demand; the volatile price makes it extremely costly to trade directly in the
ETF market, and everyone is better off investing via the OEF to smooth their idiosyncratic
shocks. The second case states that for small kǫ – which corresponds to small idiosyncratic
liquidity shocks – all investors prefer the ETF structure. The reason is that the benefit of
liquidity insurance provided by the OEF structure is too low to cover the moral hazard cost
of inefficiently unloading the aggregate shocks. We leave the discussion of the general case
to the next section.
5 Equilibrium properties and empirical implications
We now examine the properties of the equilibrium. In particular, we discuss the relative
efficiency of OEFs and ETFs in meeting individual liquidity needs, the determinants of the
OEF tracking error, the liquidity clientele for the OEFs and ETFs, and the co-existence of
OEFs and ETFs in equilibrium. Finally, we lay out empirical implications regarding the
equilibrium size of the ETF for different underlying indexes.
5.1 Efficiency of OEFs and ETFs
The main prediction of our model is that, in absence of additional frictions, the ETF structure
is as efficient as the OEF structure.
27
Result 1. ETFs are not more (or less) efficient than OEFs.
For ETFs, only traders bear their own trading costs; for OEFs, all investors might share
the trading costs of a few traders. The source of the costs is the differential price between
the price traders receive from the fund (P F1 ) and the price the fund is able to execute the
order in the underlying stock market (P E1 ). Despite of this price discrepancy, it’s a zero-sum
game between those who cause the fund flow and those who bear the flow-induced trading
costs.
This result runs directly counter to Kranefuss’s claim that ETF is a better product than
OEF since long-term traders do not subsidize the cost of short term traders. Instead, we
show that the OEF structure can be viewed as providing partial insurance against future
liquidity needs: Investors pay the cost of lower average returns in exchange for better prices
when they experience liquidity shocks. This is similar to the insurance feature of bank
deposit contracts in Diamond and Dybvig (1983).
As long as investors are risk averse, the OEF structure is actually beneficial rather than
costly to investors. However, this benefit may not be obvious to the investors, since the
reported performance of OEFs can be lower than that of ETFs due to the difference in the
way returns are accounted for. In particular, the flow-induced transaction costs are equally
shared by all investors in the OEF and reduce reported OEF performance, whereas they are
incurred only by those ETF investors with liquidity needs and do not affect reported ETF
performance.
However, the OEF structure can induce additional inefficiency. Namely, it can cause
moral hazard issues that induce excessive trading and reduce the OEF performance. In
particular, since OEF investors transact at a price that does not fully incorporate the impact
of their own trading decisions, they do not internalize the price impact of their orders. As a
result, they trade too aggressively given their trading needs, leading to excess trading costs
for the OEF. This extra trading cost makes the liquidity insurance an ex-ante negative-sum
game and leads to a lower average return for the OEF.
Result 2. The OEF structure can induce moral-hazard, i.e, excessive trading, and hence,
make OEF less efficient than the ETF
• In ETF, given P1(Y ), investors bear Y risk and trade only ǫi
• In OEF, given P F1 , investors trade both Y and ǫi
28
• When Y large, can manifest into a “bank run” for OEF
5.2 Liquidity clientele
Result 3. Liquidity clientele: investors with high liquidity needs (short-term investors) prefer
OEFs and investors with low liquidity needs (long-term investors) prefer ETFs.
• All OEF investors share the cost of liquidity insurance
• For investors with high liquidity needs (short-term investors), the liquidity insurance
provided by OEFS is more valuable
=⇒ prefer OEFs
• For investors with low liquidity needs (long-term investors), the insurance benefit is
lower than the cost of insurance
=⇒ prefer ETFs
• (Is Bogle wrong?)
• Potential Concerns:
– In practice: OEFs have short-term trading restrictions
– In theory: Does the equilibrium unravel? Would all but the highest liquidity need
investors prefer ETFs?
Third, the tradeoff between the liquidity-insurance benefit and the moral hazard cost
differs for investors with different individual liquidity needs. Under the OEF structure, the
moral hazard cost of excessive trading is shared evenly among all investors. Higher-liquidity-
need investors benefit more from the liquidity insurance and prefer to invest via the OEFs,
whereas lower-liquidity-need investors benefit less from the liquidity insurance and prefer to
invest via the ETFs. As intuitive as it may seem, this result runs directly counter to John
Bogle’s critique that “ETFs can only be described as short-term speculation.” Instead, we
find that long-term investors may optimally choose to invest in the ETFs and that there is
a viable role for ETFs in the mutual fund industry.
Result 4. OEFs and ETFs can co-exist in equilibrium
29
One might be concerned about the viability of the OEF structure given the concentration
of higher-liquidity-need investors in it. Our fourth result states that the OEF structure is
still viable and that OEFs and ETFs coexist in equilibrium with different liquidity clienteles.
The reason behind this rather surprising result is that flow-induced trading costs depend only
on the aggregate liquidity need at the fund level. Since individual liquidity needs cancel out
across investors, the concentration of investors with high individual liquidity needs does not
lead to higher aggregate liquidity need or higher flow-induced trading costs for the OEF.
This result highlights a potential pitfall of the recent trend in the mutual fund industry
to impose frequent-trading restrictions. Since higher frequency traders do not necessarily
impose higher costs to the OEF yet they benefit much more from the liquidity insurance
provided by the OEF structure, they are the best clientele for OEFs. If they are deterred
from investing in the OEFs by frequent trading restrictions, OEFs may lose a significant
client base.14
• High-liquidity-need investors enjoy the liquidity insurance, but do not impose higher
costs to OEFs
– High-liquidity-need have more volatile ǫi, i.e., high σǫ
– Given high σǫ, they prefer safer prices P F1 (background risk)
– ǫi cancels at the fund level =⇒ no additional trading costs
• No need to impose short-term trading restrictions in OEFs
• One caveat – this result does not hold if high-liquidity-need investors have higher ag-
gregate risk exposure
– For example, “sentiment” or “style” investors, or informed investors
14This result is based on the assumption that higher-liquidity-need investors do not have higher exposureto the systematic liquidity risk. If there is reason to believe that frequent traders are more likely to bereacting to market conditions or are more correlated in their trading decisions, then they may impose highercosts to OEFs and some forms of frequent trading restrictions may be optimal.
30
5.3 The tracking error of the OEF
The tracking error of the OEF is defined in Proposition 3. From (13), ∆ = ∆2Y2 > 0, hence
the tracking error always reduces the terminal payoff for the OEF. This is consistent with
the finding in the literature that fund flows are costly for the OEFs and can reduce fund
performance.
Interestingly, from (13), the coefficient ∆2 ≡ (1 + η)γσ2v/θ is strictly positive even when
η, the relative population of OEF investors, is zero. In this case, given their tiny population
weight, OEF investors do not affect the demand of the ETF investors in (15b) or the equi-
librium price P E1 in (14b). Yet there is still a price discrepancy between P E
1 (which depends
on Y ) and P F1 (which is independent of Y ). In particular, if Y > 0, the aggregate liquid
shock is equivalent to an excess supply, which reduces P E1 to below P F
1 . At the same time,
the OEF investors unload all their Y risk by selling Y shares of their stocks. Since they
are able to sell their shares to the OEF at a price (P F1 ) higher than the OEF can realize in
the stock market later (P E1 ), the OEF is losing money on average and the tracking error is
negative. Similarly, if Y < 0, OEF investors buy shares from the OEF at a price lower than
the OEF can purchase in the stock market later, also leading to a negative tracking error.
When η > 0, OEF investors follow similar strategy and unload all their Y risk. Given
their non-trivial population size, they now affect the equilibrium demand of ETF investors
and the price P E1 . In particular, each ETF investor purchases η shares for each share sold by
an OEF investor. This extra supply (ηY on a per capita basis), further depresses the ETF
price P E1 (by an extra γσ2
v ηY ). With P F0 staying constant, the OEF now loses an average
of γσ2v(1 + η)Y per share on all the Y shares sold, leading to the tracking error ∆2Y
2. This
tracking error term further reduces the demand for OEF shares and leads to additional price
impact in the ETF market, contributing to the last terms in the expressions of (14b) and
(15). Note that for small µ, η can be very large as η approaches 0. This represents the case
when the market is dominated by OEF investors. With very few ETF and stock investors
to make markets, the price P E1 can be extremely sensitive to the aggregate liquidity shock.
The OEF needs to pay a significant cost to unload all their Y risk to the small underlying
stock market, leading to significant price impacts and tracking errors.
In summary, the tracking error consists of two components. The first component is due
to the price discrepancy between the OEF price and the price in the ETF market when we
take the ETF price as exogenous. This is the moral hazard cost of providing the liquidity
31
insurance. The second component of the tracking error is a “feedback effect,” driven by the
price impact in the underlying stock market when OEF investors unload their aggregate risk
exposure. The price impact then leads to additional tracking errors for the OEF, further
decreasing investors’ demand for the OEF.15 This component of tracking error increases with
the size of the OEF. Thus, we have the following result.
Result 5. The tracking error always reduces the OEF performance, and it is not zero even
when the size of the OEF approaches zero. Moreover, the magnitude of the tracking error
increases as the relative size of the OEF (η) increases.
5.4 The cross-section of the OEF demand
Our framework allows us to assess the effectiveness of the OEF and the ETF structures
under different circumstances. Viewing the underlying risky asset in the model as a stock
index, we can compare characteristics of different stock indexes and ask whether one index
is more suited for the OEF or the ETF investors. In Figure 3, we report the equilibrium size
of the ETF relative to the OEF for different indexes and investor characteristics.
(a) Aggregate Liquidity Shock (σy) (b) Individual Liquidity Shock (σǫ)
0. 0.05 0.1 0.15 0.2ΣY
0.7
0.75
0.8
Η
0. 0.5 1. 1.5 2.ΣΕ
0.2
0.4
0.6
0.8
1.
Η
Figure 3: The equilibrium size of the ETF industry (η). Panels (a) and (b) report theoptimal size of ETF for different values of σy and σǫ, respectively. Other than the variablesbeing changed, the parameters are γ = 4, θ = 1, V = 0, Y = 0, σv = 0.3, σy = 0.1, σǫ = 1.0,and µ = 1.
First, we consider the impact of aggregate liquidity shock. A more volatile aggregate
liquidity shock (higher σy) corresponds to the case in which investors have more correlated
liquidity needs. We have the following result:
15In the paper we only consider the case of small Y for tractability. In practice, when Y is reasonablylarge, this effect can manifest into a “bank run” situation, in which the remaining investors are exceedinglyconcerned about this negative externality and may choose to sell shares even when they personally do notexperience liquidity shocks. This situation is considered in Chen, Goldstein, and Jiang (2007).
32
Result 6. When the aggregate liquidity shock is more volatile, it is more costly to provide
liquidity insurance via the OEF structure, and the equilibrium size of the ETF is larger.
With more volatile aggregate liquidity shock, the OEF is expected to have larger un-
balanced demand from its investors. This unbalanced demand is going to translate into
large excess demand in the underlying stock market, leading to a large price impact and a
large tracking error. As a result, the OEF becomes less attractive relative to the ETF as an
indexing vehicle, and the equilibrium ETF size is larger.
Empirically, it is reasonable to assume that investors of a narrower index (such as a bio-
tech index) have more correlated liquidity needs compared to investors of a broader market
index (such as the S&P 500 index). Thus, we expect to have more ETFs in narrower indexes
than in broader market indexes. We proxy for the narrowness of an index using the number
of stocks, the market capitalization, and the industry concentration of the index.
Next, we consider the impact of idiosyncratic liquidity shocks. A more volatile idiosyn-
cratic liquidity shock (higher σǫ) corresponds to the case in which investors on average have
large liquidity needs but these liquidity needs are not correlated. We have the following
result:
Result 7. When the individual liquidity shock is more volatile, the liquidity insurance pro-
vided by the OEF structure is more beneficial, and the equilibrium size of the ETF is smaller.
• ETFs are more beneficial for narrower indexes
– Investors have more correlated demand =⇒ high σy
=⇒ liquidity insurance too costly =⇒ prefer ETFs
• higher stock volatility ⇒ riskier indexes
• ETFs are more beneficial for retirement or institutional accounts
– Investors have lower idiosyncratic shocks =⇒ low σǫ
=⇒ liquidity insurance not valuable =⇒ prefer ETFs
Figure 3(b) indicates that if all investors in the economy face relatively low idiosyncratic
shocks (low σǫ), the equilibrium size of ETF can reach 1, or there is no need for the OEF. The
33
reason is that investors with low idiosyncratic liquidity demand can still generate significant
tracking errors as long as they are exposed to the aggregate liquidity shock. On the other
hand, the liquidity insurance feature – the key benefit of the OEF structure – is not very
important for them. As a result, the ETF structure dominates as an indexing vehicle.
This result points out a potential pitfall of the recent trend in the mutual fund industry
to impose trading restrictions to deter high frequency trading in the OEFs. Our findings
suggest that this strategy may not be efficient since high frequency traders do not pose high
costs to the fund, unless there is reason to believe that their trading is more correlated
with the aggregate shocks. On the other hand, those investors benefit much more from the
liquidity insurance provided by the OEF structure and hence are the natural demanders of
the OEFs. If they are deterred from investing in the OEFs by the trading restrictions, then
OEFs can potentially lose a significant client base.
Empirically, we expect investors to use more liquid assets to meet their individual liquidity
needs, hence, those indexes are likely to have a client base with larger idiosyncratic liquidity
needs. Thus, Result 7 suggests that less liquid underlying indexes should have a client base
with smaller idiosyncratic liquidity needs and the equilibrium size of the ETF is larger.
In addition, we can proxy for the idiosyncratic liquidity needs (σǫ) using the character-
istics of fund investors. For example, investors of retirement accounts or other institutional
accounts generally have lower liquidity needs than the average individual investors. Our
model suggests they may be better candidates for ETFs since they do not highly value the
liquidity insurance of OEFs.
Other comparative statistics: σv, µ, γ
In summary, we find that narrower indexes (e.g., indexes with fewer stocks or are more
concentrated) and less liquid indexes are better suited for ETFs. Hence, we predict that
these are the indexes in which more ETFs are introduced and, once introduced, ETFs enjoy
a higher growth rate. In contrast, the corresponding OEFs should experience a bigger loss
of market share to ETFs once these ETF are introduced.
6 Conclusions
Fifteen years ago, investors seeking to invest their money by tracking an index could only do
so by buying an OEF. In the past decade this paradigm has changed with the introduction
of ETFs that are similar to OEFs but are traded on the stock exchange. The huge success
34
of ETFs, seemingly at the expense of OEFs, raises a very interesting question regarding the
advantage and disadvantage of the organizational form of an OEF from the investor’s point
of view.
To address this question, we develop a theoretical model of equilibrium choice between
these two vehicles. We show that OEFs provide cross-subsidization among investors by
sharing the transaction costs for those investors experiencing large liquidity shocks, at the
cost of lower average returns for the remaining investors. While it is a zero-sum game,
the OEF structure provides partial insurance against future liquidity needs and is ex-ante
beneficial for risk averse investors. However, the insurance feature embedded in the OEF
structure can cause moral hazard issues in the form of excessive trading and increase the cost
of the insurance. We also find that investors with higher liquidity needs benefit more from the
liquidity insurance and hence prefer to invest via the OEF. Interestingly, the concentration
of high-liquidity-need investors in the OEF does not lead to higher flow-induced trading
costs since individual liquidity needs cancel out at the fund level. As a result, OEFs are
not dominated by ETFs in equilibrium. Moreover, we predict that the ETF is a more
suitable investment vehicle when investors have more correlated liquidity shocks or when the
underlying indexes are narrower or less liquid.
This paper is the first seeking to understand the growth of the ETF industry. Our main
finding is that although this industry is growing and will probably continue to grow in the
future, it is not likely to totally replace the standard OEFs. ETFs are a new and exciting
addition to the different investment options available both to individual and institutional
investors. However, our model demonstrates that one cannot view all ETFs as equal. Though
some ETFs are very similar to OEFs and offer investors a fairly identical investment vehicle
(except for the liquidity cross-subsidization), other ETFs offer new options to track narrower
and less liquid indexes that are not cost effective in the OEF structure.
35
References
Agapova, A., 2006, “Innovations in Financial Products. Conventional Mutual Funds versusExchange Traded Funds,” Florida Atlantic University Working Paper.
Amihud, Y., 2002, “Illiquidity and Stock Returns: Cross-Section and Time-Series Effects,”The Journal Financial Markets, 5, 31–56.
Berk, J. B., and R. C. Green, 2004, “Mutual Fund Flows and Performance in RationalMarkets,” Journal of Political Economy, 112, 1269–1295.
Berk, J. B., and R. Stanton, 2007, “Managerial Ability, Compensation, and the Closed-EndFund Discount,” The Journal of Finance, 62, 529–556.
Chalmers, J. M. R., R. M. Edelen, and G. B. Kadlec, 2001, “On the Perils of IntermediariesSetting Security Prices: The Mutual Fund Wildcard Option,” Journal of Finance, 61,2209–2236.
Chay, J., and C. A. Trzcinka, 1999, “Managerial performance and the cross-sectional pricingof closed-end funds,” Journal of Financial Economics, 52, 379–408.
Chen, J., H. Hong, M. Huang, and J. D. Kubik, 2004, “Does Fund Size Erode Mutual FundPerformance? The Role of Liquidity and Organization,” American Economic Review, pp.1276–1302.
Chen, Q., I. Goldstein, and W. Jiang, 2007, “Payoff Complementarities and FinancialFragility: Evidence from Mutual Fund Outflows,” Working Paper.
Cherkes, M., J. Sagi, and R. Stanton, 2007, “A Liquidity-Based Model of Closed-End Funds,”The Review of Financial Studies, Forthcoming.
Chordia, T., 1996, “The Structure of Mutual Fund Charges,” Journal of Financial Eco-nomics, 41, 3–39.
Christoffersen, S. E. K., D. B. Keim, and D. K. Musto, 2007, “Valuable Information andCostly Liquidity: Evidence from Individual Mutual Fund Trades,” McGill University andUniversity of Pennsylvania Working Paper.
Christoffersen, S. E. K., and D. Musto, 2002, “Demand Curves and the Pricing of MoneyMangement,” The Review of Financial Studies, 15, 1499–1524.
Diamond, D. W., and P. H. Dybvig, 1983, “Bank Runs, Deposit Insurance and Liquidity,”Journal of Political Economy, 91, 401–419.
Diamond, D. W., and R. E. Verrecchia, 1981, “Information aggregation in a noisy rationalexpectations economy,” Journal of Financial Economics, 9, 221 – 235.
Dickson, J. M., J. B. Shoven, and C. Sialm, 2000, “Tax Externalities of Equity MutualFunds,” National Tax Journal, 53, 607–628.
Edelen, R. M., 1999, “Investor Flows and the Assessed Performance of Open-end FundManagers,” Journal of Financial Economics, 53, 439–466.
36
Edelen, R. M., R. Evans, and G. B. Kadlec, 2007, “Scale effects in mutual fund performance:The role of trading costs,” Boston College, University of Virginia, and Virginia TechWorking Paper.
Elton, E. J., M. J. Gruber, G. Comer, and K. Li, 2002, “Spiders: Where Are the Bugs?,”The Journal of Business, 75, 453–472.
Gastineau, G. L., 2004, “Protecting Fund Shareholders From Costly Share Trading,” Finan-cial Analyst Journal, May/June, 22–32.
Goetzmann, W. N., Z. Ivkovic, and G. K. Rouwenhorst, 2001, “Day Trading InternationalMutual Funds: Evidence and Policy Solutions,” The Journal of Financial and QuantitativeAnalysis, 36, 287–309.
Greene, J. T., and C. W. Hodges, 2002, “The Dilution Impact of Daily Fund Flows onOpenend Mutual Funds,” Journal of Financial Economics, 65, 131159.
Grinblatt, M., and S. Titman, 1989, “Mutual Fund Performance: An Analysis of QuarterlyPortfolio Holdings,” The Journal of Business, 62, 393–416.
Hasbrouck, J., 2006, “Trading Costs and Returns for US Equities: Estimating EffectiveCosts from Daily Data,” New York University Working Paper.
Hortacsu, A., and C. Syverson, 2004, “Product Differentiation, Search Costs, and Compe-tition in the Mutual Fund Industry: A Case Study of S&P 500 Index Funds,” QuarterlyJournal of Economics, 119, 403–456.
Huang, J., and J. Wang, 2008, “Market Liquidity, Asset Prices, and Welfare,” Journal ofFinancial Economics, forthcoming.
ICI, 2008, Mutual Fund Fact Book, Washington, D.C.: Investment Company Institute.
Johnson, W. T., 2004, “Predictable Investment Horizons and Wealth Transfers among Mu-tual Fund Shareholders,” Journal of Finance, 59, 19792011.
Kacperczyk, M., C. Sialm, and L. Zheng, 2007, “Unobserved Actions of Equity MutualFunds,” The Review of Financial Studies, Forthcoming.
Lee, C. M. C., A. Shleifer, and R. H. Thaler, 1991, “Investor Sentiment and the Closed-EndFund Puzzle,” The Journal of Finance, 46, 75–109.
Massa, M., 1997, “Why So Many Mutual Funds? Mutual Funds, Market Segmentation andFinancial Performance,” INSEAD and CEPR Working Paper.
Nanda, V., M. P. Narayanan, and V. A. Warther, 2000, “Liquidity, Investment Ability, andMutual Fund Structure,” Journal of Financial Economics, 57, 417–443.
Pontiff, J., 1996, “Costly Arbitrage: Evidence from Closed-End Funds,” The QuarterlyJournal of Economics, 111, 1135–1151.
Poterba, J. M., and J. B. Shoven, 2002, “Exchange-Traded Funds: A New Investment Optionfor Taxable Investors,” American Economic Review, 92, 422–427.
37
Shleifer, A., and R. W. Vishny, 1997, “The Limits of Arbitrage,” The Journal of Finance,52, 35–55.
Stein, J. C., 2005, “Why are most Funds Open-end? Competition and the Limits of Arbi-trage,” The Quarterly Journal of Economics, 120, 247–272.
Svetina, M., and S. Wahal, 2008, “Exchange Traded Funds: Performance and Competition,”Arizona State University, Working Paper.
Viceira, L. M., and A. B. Wagonfeld, 2007, “Barclays Global Investors and Exchange TradedFunds,” HBS Cases, N9-208-033.
Zitzewitz, E., 2006, “How Widespread was Late Trading in Mutual Funds?,” AmericanEconomic Review, 96, 284–289.
38
A Appendix
Proof of Proposition 1
Plugging P E2 = V into the definition of terminal wealth W E
i2 for an ETF investor in (8a) and
integrating over the distribution of V at given Yi and θEi1, we have
E[W Ei2 | Yi] = θP E
1 + θEi1(V − P E
1 ) (A1a)
Var[W Ei2 | Yi] = (θE
i1 + Yi)2σ2
v (A1b)
Thus, the expected utility JE1 at time 1+, defined as the utility (5) conditional on Yi is:
JE1 ≡ max
θEi1
[
θP E1 + θE
i1(V − P E1 ) −
1
2γσ2
v(Yi + θEi1)
2
]
. (A2)
The optimal holding is calculated by solving the first-order condition with respect to θEi1,
θEi1 =
V − P E1
γσ2v
− Yi. (A3)
The holdings for stock investors are identical to those for ETF investors. Plugging Yi = Y +ǫi
and (A3) into the market clearing condition, from (4), we have
∫
θEi1 +
∫
θSi1 = (η + µ)
(
V − P E1
γσ2v
− Y
)
= (η + µ)θ + XF . (A4)
Solving (A4) yields the equilibrium price P E1 in (9). The optimal holding in the proposition
is obtained by substituting the equilibrium price P E1 back into (A3).
Proof of Proposition 2
The proof of (12a) is almost identical to that of Proposition 1, except that E[P F2 ] = V −∆.
We then plug the definition of ∆ into individual holding (12a) and aggregate over all OEF
investors:
∫
i
θFi1 =
∫
i
(
V − ∆ − P F1
γσ2v
− Y − ǫi
)
= (1 − η)
(
V − ∆0 − ∆1Y − ∆2Y2 − P F
1
γσ2v
− Y
)
.
Imposing the restriction that∫
iθF
i1(Y = 0) = (1 − η)θ yields the expression for P F1 in the
proposition. Substituting this P F1 back into (12a) to derive the expression in (12b).
39
Proof of Proposition 3
To calculate the tracking error, we plug the expressions of P E1 , P F
1 and θFi1 from Propositions 1
and 2 into the definitions of XF and ∆ in (6) and (7). Since∫
iǫi = 0, we have
∆ =((1−η)θ−
∫
iθF
i1)(PF1 −P E
1 )∫
iθF
i1
(A5a)
=Y (γσ2
v + ∆1 + ∆2Y )(−∆0 + ((1 + η)γσ2v + η(∆1 + ∆2Y ))Y )
γσ2v(θ − Y ) − (∆1 + ∆2Y )Y
(A5b)
When Y is small, we can use Taylor expansion to expand around Y and drop higher
order terms (i.e., O(Y 3)). Equalizing the coefficients of zero, first, and second order terms
in (A5b) to those in the definition of ∆ ≡ ∆0 + ∆1Y + ∆2Y2, we get three equations and
three unknowns (∆0, ∆1, and ∆2). The solution is ∆0 = ∆1 = 0 and ∆2 is given in (13).
Parts (ii) and (iii) are straightforward. Substituting the expressions of ∆0, ∆1, and ∆2
into (11) and (12) yields P F1 and θF
i1. Using the definitions of ∆, XF in (6) and Proposition 1,
we derive the equilibrium expressions of P E1 and θE
i1.
Proof of Lemma 1
First, we calculate the time 0 expectation and variance of W Fi2 conditional on the realization
of Y .
E[W Fi2 | Y ] = E[E[W F
i2 | ǫi, Y ]] = θV − Y (γσ2
v + 2∆2Y )θ + O(Y 3)
Var[W Fi2 | Y ] = E[Var[W F
i2 | ǫi, Y ]] + Var[E[W Fi2 | ǫi, Y ]]
= (1 + s2
i kǫ)θ(σ2
v θ − 2∆2Y2/γ) + O(Y 3)
We then calculate the unconditional expectation and variance:
E[W Fi2 ] = E[E[W F
i2 | Y ]] = θV − 2θ∆2σ2
y + O(σ4
y)
Var[W Fi2 ] = E[Var[W F
i2 | Y ]] + Var[E[W Fi2 | Y ]]
= (1 + s2
i kǫ + ky)kθ/γ2 − 2(1 + s2
i kǫ)θ∆2σ2
y/γ + O(σ4
y)
Then, JFi0 = E[W F
i2 ] − (γ/2)Var[W Fi2 ] yields (16a).
Similarly, we calculate the time 0 expectation and variance of W Ei2 conditional on the
40
realization of Y .
E[W Ei2 | Y ] = E[E[W E
i2 | ǫi, Y ]] = θV + γσ2
v(θx1Y + (x1 + x2
1 + θx2)Y2) + O(Y 3)
Var[W Ei2 | Y ] = E[Var[W E
i2 | ǫi, Y ]] + Var[E[W Ei2 | ǫi, Y ]]
= (1 + s2
i kǫ)σ2
v(θ2 + 2θ(1 + x1)Y + ((1 + x1)
2 + 2θx2)Y2) + O(Y 3)
We then calculate the unconditional expectation and variance:
E[W Ei2 ] = E[E[W E
i2 | Y ]] = θV + ky(x1 + x2
1 + x2θ)/γ + O(σ4
y)
Var[W Ei2 ] = E[Var[W E
i2 | Y ]] + Var[E[W Ei2 | Y ]]
= (x2
1kykθ + (1 + s2
i kǫ)(kθ + ky((1 + x1)2 + 2x2θ)))/γ
2 + O(σ4
y)
Then, JEi0 = E[W E
i2 ] − (γ/2)Var[W Ei2 ] yields (16b).
Proof of Proposition 4
By substituting the definitions of ∆2, x1, and x2 from Proposition 3 into the value functions
JFi0 and JE
i0 in Lemma 1, we derive
JFi0 = θV −
kθ
2γ
(
1+s2
i kǫ+ky
)
−1+η
γ(1−s2
i kǫ) ky + O(σ4
y)
JEi0 = θV −
kθ
2γ
(
1+s2
i kǫ+ky
)
−(1+η)(1+3η)
2γs2
i kǫ ky −1−η2
2γ(1−kθ) ky + O(σ4
y).
Taking the difference between JFi0 and JE
i0 yields (17). Clearly, an individual investor should
invest in OEF if and only if GFi ≥ 0.
To prove part (ii), we take derivative of GFi :
∂GFi
∂si
=(1 + η)2
γ3sikǫky > 0.
Proof of Proposition 5 and Corollary 1
We have proved the cases of GFi (0, 0) > 0 and GF
i (1, 1) < 0 of Proposition 5 in the text.
Using (17), we can verify that the first two conditions in (19) correspond exactly to those
two conditions.
We now consider the case of GFi (0, 0) ≤ 0 ≤ GF
i (1, 1). Whenever kǫ ≥ (1 + kθ)/3, we can
verify that the expression of η in the third case of (19) is bounded between (0, 1) and solves
GFi (η, η) = 0.
41
For any si > η, we have GFi (si, η) ≥ GF
i (η, η) = 0, hence si invests via the OEF. For
any si < η, we have GFi (si, η) ≤ GF
i (η, η) = 0, hence si invests via the ETF. Given that
si ∈ Unif [0, 1], the population of investors with si < η is exactly η. Hence, the population
of ETF investors is η.
42
Table 1: Summary Statistics - Time Trend
This table provides descriptive statistics of the full sample of OEFs and ETFs for the years 1992 to 2006. The data is basedin data collected from: CRSP’s Survivor-Bias Free U.S. Mutual Funds Database, EDGAR’s SEC Filings and Forms Database,Bloomberg, textitThomson Financials’s CDA/Spectrum Mutual Funds Holding, and CRSP. OEFs are Open-Ended IndexMutual Funds. All OEF share classes are collapsed to one entity. ETFs are Exchange Traded Funds tracking indexes andtrading on the secondary market. All sizes are in millions of dollars. The percentage growth numbers are winsorized at the 2%level.
Year 1992 1996 2001 2006Variable OEF OEF ETF All OEF ETF All OEF ETF AllNumber of Funds 62 81 19 100 182 110 292 200 320 520Number of Families 30 40 2 42 79 8 86 72 23 92Number of Indexes 15 21 19 38 32 108 132 55 268 283
Funds per Family Mean 1.90 2.00 9.50 2.36 2.30 13.75 3.40 2.78 13.87 5.64Median 1 1 10 2 1 4 1 1 4 2Std 1.81 1.89 10.61 2.96 2.45 24.56 8.17 4.28 25.21 14.50
Number of Indexes Mean 1.76 1.76 9.50 2.19 2.03 13.75 3.26 2.60 13.70 5.22per Family Median 1 1 10 2 1 4 1 1 6 2
Std 1.55 1.58 10.61 2.96 1.94 24.56 8.45 3.91 23.75 13.03Fund Size ($M) Mean 281.84 1,037.22 49.91 849.63 1,655.36 436.00 1,196.02 3,077.35 1,306.63 1,988.99
Median 88.42 259.64 6.00 189.23 195.59 69.33 121.01 386.59 187.48 268.35Std 860.69 3,608.74 159.02 3,267.99 7,315.70 1,718.16 5,894.47 11,536.84 4,756.05 8,109.44
Family Size ($M) Mean 575.63 2,100.36 474.14 2,022.93 3,813.62 5,995.03 4,060.89 8,548.20 18,122.43 11,220.50Median 179.36 317.90 474.14 317.90 431.48 3,086.48 468.67 952.95 1,996.18 1,091.17Std 1,780.78 8,731.34 464.77 8,523.24 21,674.23 7,814.93 21,001.50 50,520.57 50,000.55 53,017.65
Percentage Mean 57.00 64.08 277.31 67.72 7.40 389.11 106.53 27.71 107.78 19.00Fund Growth Median 39.17 56.44 277.31 56.44 -3.62 58.08 -0.68 16.51 54.11 19.00
Std 92.69 56.58 357.10 65.74 43.51 1,026.96 305.29 45.03 150.76 19.00Percentage Mean 68.06 68.60 39.19 67.71 -3.57 121.63 11.42 16.11 93.30 27.40Family Growth Median 51.00 60.07 39.19 59.30 -7.89 75.31 -4.43 9.81 46.12 15.33
Std 96.80 71.52 0.00 70.58 23.42 116.30 64.31 31.16 130.15 50.94
43
Table 2: Summary Statistics of Underlying Indexes Tracked
This table provides descriptive statistics of the indexes tracked by the ETFs and OEFs inthe sample. The numbers are averages across all indexes over the entire time of the sample,1992-2006. The indexes are sampled every quarter and the changes are recorded from onequarter to the other. The liquidity measure is based on Amihud (2002). Amihud’s LiquidityMeasure (l1) is the absolute value of the return of a stock divided by the absolute value ofits price multiplied by its volume. We calculate the liquidity measure of each stock in theindex using data downloaded from Joel Hasbrouck’s website and described in Hasbrouck(2006) and average it across the index either Equally-Weighted (EW) or Value-Weighted(VW) by the market cap of the stock. We use two industry concentration measures. First,we use the Herfindahl-Hirschman Index using either the Fama-French 10 or 48 industriesclassifications. We assign to each stock in the index an industry based on the firm’s SICcode. Second, we use the industry concentration following Kacperczyk, Sialm, and Zheng(2007).
Variable OEF ETF AllIndex Size:
- Number of Stocks in Index 593.66 325.73 487.41- Market Cap of Index (in $B) 5656.22 2337.15 4443.42Index Turnover (over a quarter):- Number of Stocks Changing in the Index 13.60 7.47 11.22- Percentage of the Stocks in Index Changing 2.49% 3.19% 2.74%- Market Cap Changing in the Index (in $B) 378.40 159.23 299.02- Percentage of Market Cap in Index Changing 10.22% 18.84% 13.34%Liquidity of Underlying Stocks:
- Amihud Liquidity Measure (EW) 0.031 0.074 0.060- Amihud Liquidity Measure (VW) 0.003 0.022 0.016Industry Concentration:
- Herfindahl Index (10 industries) 0.23 0.47 0.32- Herfindahl Index (48 industries) 0.13 0.32 0.20- KCZ Industry Concentration 0.09 0.31 0.17
44